Grok Bot pricing: $40 to $300 a month, and the meter nobody caps

How much does Grok Bot cost?

Grok Bot has no standalone price. It is a benefit attached to a subscription you buy for something else, and there are six of them. The get-started documentation names the eligible set precisely: "SuperGrok Plus, SuperGrok Heavy, Cursor Pro+, Cursor Ultra, or Cursor Teams Standard or Premium". Cursor’s own pricing page independently confirms Grok Bot on Pro+, Ultra, Teams Standard, and Teams Premium.

One trap sits inside the SuperGrok ladder specifically. Base SuperGrok at $30 a month does not include Grok Bot. It includes Grok Build, the coding CLI, which is a different product with a similar name. Grok Bot starts at SuperGrok Plus, two tiers higher and $70 more. Anyone who upgraded to SuperGrok expecting Bots has bought the wrong thing.

A $99 a month promotional price for SuperGrok Heavy has also been reported, with sources disagreeing on whether it runs for three months or six. That is a $603 difference in the first year, the reporting conflicts, and we could not verify it against xAI. Treat it as a rumour to check at checkout rather than a price to plan on.

The practical read: if you already hold Cursor Pro at $20, you are one tier short and the upgrade to Pro+ is the cheapest individual route. If you are buying for a team, a Teams Standard seat at $40 is the least expensive documented way in and carries Grok Bot for every member. If you hold SuperGrok Heavy for the chatbot, you already have it. Cursor’s plans take roughly 20 percent off when paid annually, which is the only discount lever available on any of these paths.

What changed and when

Grok Bot pricing moved twice over 11 calendar dates, and the second move invalidated the entire ranking set for this query. This is the dated record.

One consequence is worth stating flatly, because it will mislead anyone comparing pages. The most rigorous competitor article on this topic, published on launch day, explicitly marks Cursor Pro+ as "not an eligible Grok Bot tier". That was correct when written and is wrong now. Another states a price floor of $120 to $300 a month, which the 21 August expansion cut to $40 for a team seat. If a page you are reading does not carry a date after 21 August 2026, its tier list is stale.

The link between your accounts is permanent

This is the most consequential thing on this page and no article covering Grok Bot pricing mentions it. Because every access path requires a Cursor account, using Grok Bot means linking your Grok account to a Cursor account. Cursor’s own documentation states what that link is:

Read that as a purchasing decision rather than a technical note. The moment you link, you have made an irreversible choice about which Cursor account your Grok subscription is attached to, forever. There is no undo, no migration path, and no support process documented for changing your mind.

• Link your personal Grok to a work Cursor account and your personal subscription is permanently tied to an account your employer controls. If you leave, the link does not.
• Link to the wrong account in a hurry and you cannot move it. The remedy is a new Grok account and a new subscription.
• Restructure your team’s Cursor org later and existing SuperGrok links cannot follow. This is worth raising before a migration, not after.
• Evaluate on a throwaway Cursor account and you have burned that link for the Grok account you used, permanently.

The practical advice is short: decide which Cursor account should own this before you click anything, and do the evaluation with a Grok account you are willing to strand. An irreversible commitment that nobody warns buyers about is exactly the kind of thing that is cheap to get right on day one and impossible to fix on day thirty.

Is there a Grok Bot free trial, and how long is it?

A trial exists. Its length is genuinely unknown, and the sources that claim to know disagree with each other.

What xAI actually said, through its official @bot account on 21 August 2026, is this: "We’re also offering a free trial with limited usage for all other users." The words "limited usage" appear; a number of days does not. The x.ai announcement page covering the same expansion lists the plan changes and does not mention the trial at all. No duration is stated on the product page, the announcement, or any of the Grok Bot documentation pages.

The meter nobody caps

The seat price is the part of Grok Bot pricing everybody writes about. The meter underneath it is the part that decides what you actually pay, and it has four properties that compound badly together.

• The weekly token allowance is not published. You are told there is one. You are not told how large it is, in tokens, in hours, or in tasks.
• Past it, overage is billed at raw model rates. Not a discounted bundle rate, the underlying per-token price.
• There is no Grok Bot spend cap. As eesel’s teardown put it, "There is no Grok Bot-specific spend cap yet." Nothing stops a looping Bot at any dollar figure.
• There is no model picker, so you cannot route inexpensive work to an inexpensive model. The router decides, and it does not tell you what it decided.

To size the risk, use the only rates xAI publishes for the model that is plausibly doing the work. Grok 4.6 lists at $2 per million input tokens and $6 per million output for 200K prompt tokens or fewer, and $4 and $12 above 200K. Note the cliff: crossing above 200K tokens in a single prompt doubles the price of the entire request, and an always-on Bot with compounding memory and a persistent filesystem is exactly the workload that drifts toward long prompts over time.

The four layers of the real bill

A budget built on the seat price alone will be wrong. There are four layers, and only the first is on any pricing page.

The fourth layer is the one people underestimate, because it is the only one that gets worse the more capable the agent becomes. An agent that can only draft is safe to test. An agent that can operate your accounts, on a machine holding all your logins, with no rehearsal mode and no undo, has a failure mode whose cost is bounded by what your accounts can do, not by your subscription.

Four realistic monthly configurations

Nobody buys a subscription and then runs nothing. These are four shapes a real month takes, with the seat priced from the sources above and the meter modelled rather than quoted, because xAI publishes neither the allowance nor a usage-to-cost mapping. Treat the seat column as fact and the total column as a planning estimate.

The bottom row is the honest one and it is why this page exists. Every other row is a subscription with a number on it. That row is a subscription plus a meter with no ceiling, feeding a router you cannot steer, measured by a display that did not work at launch. There is no arithmetic that turns it into a number, and anyone publishing one is guessing.

Two structural points change the sums more than any individual price. First, always-on is the expensive mode by construction. The reason to buy this product is agents that keep working while you are away, and that is precisely the usage pattern that consumes an allowance fastest. The feature you are paying for is the feature that makes you exceed the plan. Second, a subscription fails safe and a meter fails expensive. When a plan window closes, work stops and you notice. When a metered agent loops overnight against a task it cannot complete, nothing stops it and you find out on the invoice. Grok Bot today is the second kind with no cap available, and launch-week reports describe Bots looping in exactly that way.

What the same work costs on hardware you already own

Grok Bot rents you a computer. The alternative shape is to run the agents on a machine you already have and pay only for inference. It is worth pricing the two honestly, including where the rented computer genuinely wins.

The rented computer wins on two things and they are not small. There is nothing to set up: no machine to keep awake, no networking, no CLI to install and authenticate. And the machine is somebody else’s problem when it breaks. If your objection to running agents locally is "I do not want to babysit a computer", that objection is correct and Grok Bot answers it.

What it charges for answering it is the rest of this page: an unpublished allowance, an uncapped meter, a router with no picker, a shared machine with no boundary between agents, and a single vendor for both the agent and the model.

The two bottom rows are the honest concession, and anyone telling you otherwise is selling. If nobody at your desk will keep a machine awake, a rented computer is the answer and this comparison ends there.

What the price does not include

Four things a buyer reasonably assumes come with a $200 or $300 subscription, and which do not come with this one as of 22 August 2026.

• An enterprise agreement. Enterprise is a waitlist, not a plan. There is no seat count at which you can buy a contract, a DPA, or a support obligation. Whatever tier you are on, you are on the self-serve product.
• An API. No public API and no webhooks. Nothing you build can start a Bot or read its output, so the subscription buys a product, not a capability you can compose with.
• A predictable ceiling. There is no spend cap to set, no published allowance to plan against, and no model picker to control what a task costs. You are buying access to a meter, not a budget.
• Independence from Cursor. Access, identity, billing, and the governing terms all run through Cursor. A Cursor-side policy change can remove the product: Legacy Privacy Mode blocks Grok Bot outright, at any price.

The Cursor dependency is worth pricing as a commercial risk rather than a technical footnote. You are buying an xAI product whose availability is contingent on an account with a different brand, under terms that belong to that brand, following an acquisition that had not even closed when the product shipped: Grok Bot launched on 11 August 2026 and SpaceX closed its purchase of Cursor’s maker on 14 to 15 August. The product also runs on Cursor’s infrastructure rather than xAI’s. That is not a prediction that it will go wrong. It is a reason to avoid building a critical weekly process on it before an enterprise agreement exists to sign.

Which access path is cheapest for you

Whatever you pick, do the arithmetic on the meter and not on the seat. The seat is between $40 and $300 and you know it in advance. The meter has no published allowance and no cap, and it is metered against a model you did not choose. If you want the sum done for your own usage shape, the cost calculator does it, and reducing AI costs covers the general problem.

Questions people ask

Keep reading

Continue through Grok Bot with Grok Bot review (August 2026): is the $40 to $300 agent worth it?.